feature/dueling dqn

[ryanpeach]

• Actor critic depends on dqn
• Renaming based on the implementation type
• This pre-commit hook is making commits hard
• saving work
• Going to also do ddqn because it introduces the advantage function
• Experiment: DDQN, last_10_percent_mean: [94.82, 99.26, 93.82, 96.94, 90.26, 79.8]
• Did ddqn, still need notes
• Experiment: DDQN, last_10_percent_mean: [90.02, 87.16, 98.1, 99.84, 95.58, 97.88]
• Need expirement manager to only look at dirty when its initialized
• Experiment: DDQN, last_10_percent_mean: [74.12, 88.0, 93.96, 96.52, 94.8, 95.2]
• Some renames

Great job making a new implementation!

Here are some tasks to complete before merging this PR:

Styleguide

Jupyter

• Make sure it runs on cpu and gpu
• Comment the shapes of any numpy arrays or torch tensors. Make assertions on the output.
• Functions that have more than one parameter should have a * before the first or second parameter to force the user to use named arguments, unless there is only one parameter.
• Do not set default values, you might forget to pass parameters up the stack.
• Add the proper headers to the notebook:

%load_ext autoreload
%autoreload 2

• Make sure the jupyter notebook is importable without running any code. Most cells should look like this:

def foo():
    pass

if __name__ == '__main__':
    foo()

Logseq

• Make logseq notes and flash cards.
• Do not use logseq aliases so that the graph looks clean and its more navigable.
• Use singular nouns for tags.
• Use spaces in filenames instead of - or _ just so that you don't have to use aliases (ugly I know).